The Pippa Story: an AI invention could prevent 50% of house fires!

“This data proved to us some very important things – firstly that the fires and hazards were all detectable using thermal imaging,

but most importantly that they were predictable.”

 

From the Inventor, Sam Bailey:

  • Imagine an idea that could halve the number of housefires – we have pretty much invented something that can…
  • We have called our invention “Pippa”

My team and I are the inventors of several products in the Connected Home, AI and Internet of Things sectors, among them the Skyrad solar heating system, the LeakBotTM self-fit water leak detector, the HeatDoctor heating diagnostic system, (and several others which are still in stealth mode while we protect our IP).

LeakBot has gone on to win several awards, including the Insurance Times IoT Breakthrough award. Read about it in the Economist .

Developing LeakBot has given me and the Pippa team (Ben, Lewis, William, Jamil and Mark) the experience needed to bring a new IoT invention to market.

Housefires – The Facts Are Frightening

I had been considering, based on our background in IoT, insurance and AI development, what other new products and inventions could really help make people’s lives safer and more convenient.

One of the big risks that households face is housefires. England has around 30,000 dwelling fires per year that the Fire Service have to attend to (there will be many more that the householder manages to put out, and don’t get recorded). This is 1 in every 800 households each year having a housefire. The number may seem low, but over an 80-year lifespan, that means around 1 in 10 of us will at some point experience the horror of a housefire.

Housefires are also the third most expensive risk that home insurances pay out for, after water leaks and weather damage. For the insurers we looked at it, the typical claim value for a housefire is around £50,000.

Using some statistical models we estimated the value of protecting people from those fires. We also looked at what was causing those fires. Fire service data in the UK and elsewhere shows that across developed countries, a fairly consistent 50% of housefires are started on by cooking.

Cooking fires have become dominant partly because other risks have been reduced substantially. Fewer people smoke, and the safety of electrical wiring and appliances, and the combustibility of materials used in homes have improved over time. However the one risk that hasn’t been reduced, and that technology hasn’t addressed properly yet, is the risk from cooking fires.

Research

So we started a research program – to see whether using our expertise in artificial intelligence and smart sensors would help us to develop a product that could protect people from these risks. We started out with a fairly expensive thermal imaging camera similar to the ones that the military use. This was never going to be a viable product as it was too expensive to put in every home, however we expected that the cost of the sensors would fall over time.

We created a test rig outdoors using a camping stove, and put on protective gear and intentionally created various cooking hazards – setting chip pan fires, boiling pans dry, setting fire to tea towels left near the stove, and letting food intentionally combust, all the while capturing data from a thermal imaging camera test rig. This data proved to us some very important things – firstly that the fires and hazards were all detectable using thermal imaging, but most importantly that they were predictable. We could tell that fires were going to start, several minutes before they caught fire.

Patents & Innovation – It’s very rare that an idea is completely new!

Based on these tests we developed some designs for algorithms, and we filed patents on the concept. The patents were granted with no prior art found. This is very unusual in any invention – most new products are sufficiently similar to pre-existing ideas that patent examiners will find something. This indicated to us that we have definitely found a new and innovative invention. It also gives protection from our design being copied, and the patent has now been filed worldwide.

Product Development – Prototypes

An important next step was to prove that we could make the AI work in real cooking situations, so that we could minimise the chance of a false alarm during normal cooking and cause a nuisance to users. If the device become a nuisance, users might remove it or take the batteries out. To do this we needed to get prototypes into people’s kitchens that could gather data for us as quickly as possible.

We used Raspberry Pis with miniture thermal imaging cameras, and mounted them above our stoves, and those of friends and family, uploading data to our cloud algorithms. What we could then do was train our algorithms and optimise them based on the data from real people’s houses, we could train them to detect the real hazards that we created and simulated, and not trigger when there is normal cooking usage, despite different stove types and cooking styles.

Product Design – Making Pippa Stylish & Small

In the meantime, we started designing what the final hardware and case would look like. We chose a design that has a very similar shape and size to a light switch. This makes it stylish and inconspicuous in a kitchen and helps it to blend in. We also designed how the device would be fitted, as it is very important that a user can easily and safely fit one in their own kitchen without needing tools or doing any damage to their kitchen.

We made a few important design decisions – the first was to make Pippa speak before it triggered an alarm. Most alerts from a Pippa are benign if you respond quickly – you have let your pan get too hot and your food is about to burn, or you’ve stopped cooking and not turned the stove off. Pippa only emits a high-volume alarm if you don’t respond to the gentle spoken reminders in a reasonable time.

Connectivity

We also decided to connect Pippa to our network so it can warn you if you are out of the house or don’t respond to an alarm, and can also alert a friend, neighbour or carer if there is a hazard. This is particularly important for vulnerable users or dementia sufferers, who are at very high risk of stove fires. Pippa connects via a cellular phone network, rather than Wi-Fi. This incurs a cost as we need to pay for a SIM subscription whereas Wi-Fi would be free. However there are two risks with Wi-Fi that we have found with some of our previous IoT products. The first is that many people found the set up of a Wi-Fi device very fiddly, and they tended to struggle to get things like the paring and password setup correctly. By adding a cellular connection, the Pippa can be ready to go straight out of the box. This problem is particularly acute amongst older people and those who are less tech-savvy. You can buy a Pippa for your elderly parent or grandparent, or put one in your care home or rental property and it will just connect automatically when they install it. It doesn’t need them to have Wi-Fi or a smartphone to work.

Another problem we found was that IoT devices often lose their Wi-Fi connection over time, as routers update DHCP tables, people switch ISP, move routers or change passwords etc. We’ve all had devices that mysteriously lose their Wi-Fi connection and need resetting or rebooting before you can use them again. Given a Pippa might sit on your wall for several years before you need it to warn you of a fire, we really don’t want it to have lost its connection in the interim.

Reliability & Proving Pippa Works!

We also had to test that the device was sufficiently reliable, and that it could last up to 12 months using just 2 AA batteries. We have now had Pippas in peoples kitchens for 5 years, testing and refining the algorithms, the hardware and the software. We have also set fire to many products that a Pippa might experience – frying pans, chip pans, cooking oil spills, baked beans, pasta, tea towels, paper, clothing etc. all to train the AI to recognise their different pre-fire and fire signatures. The recognition rate is extremely high – we will never be able to guarantee to predict or detect every fire, but we estimate that comfortably over 95% are detected by the Pippa algorithm. In the case of unattended stove fires such as oil fires or pans boiling dry and combusting, we can predict them several minutes before they break out. For fires that start instantaneously (for example some leaves a tea towel very close to a hot stove and doesn’t realise) we can detect it in around 30 seconds. Most smoke detectors take over 5 minutes to respond because they need either a large amount of smoke, or a very high rise in the temperature in the kitchen, which won’t typically happen until the fire has already spread and is out of control.

We also built and tested the cloud interface that could enable users to register the device and to set up remote alarms so that they can be automatically phoned or texted if Pippa detects a hazard and they are out of hearing and don’t notice the alarm – for example if they had left the stove on and left the house. We’ve built the system using the Amazon cloud, and integrated all of Amazon’s latest security technology to ensure the security of our users data.

The Amazing AI Cooking Assistant Idea & Pippa App

As well as being a scientist and stats geek, I am also a keen cook. I’d been thinking for several years about the advantages of having some sort of sensor on a stove top. Knowing the temperature of your pan would make it safer and easier to cook, and in particular it would make it easier for people who are less experienced cooks to get meals cooked at the right temperature. Gauging whether it is the right time to add your food to a frying pan, or tracking how long each dish has been simmering is very hard. Experienced cooks use subtle cues like the sound of the sizzle, or the smell coming from the pan, to judge when their food is ready. Getting it wrong in one direction means burning your food and filling your kitchen with smoke. Get it wrong the other way and you serve undercooked food, risking food poisoning, or you add food to oil before it is hot, causing it to soak the oil up before it seals resulting in unpleasant food with a higher fat content. An AI system with a sensor would help solve this, and it would also make cooking repeatable, so if I cook something once I can then cook it again. It would also make people less likely to have a fire on their stove.

So we also created the smart assistant app. The app connects your smartphone or tablet to your Pippa vie Bluetooth, to turn the Pippa into an intelligent cooking Assistant.

Some foods require cooking at quite specific temperatures in order to get the desired result – examples of these are frying a steak, cooking pancakes, poaching eggs, and for the more adventurous sous vide cooking where food is cooked a low temperature in a sealed bag in order to maintain the juiciness of the food.

In recipe mode the Pippa app is designed to guide you to the right pan temperature, to time your cooking and tell you when to add or turn your food, and when it is cooked. This way even very inexperienced cooks can get perfect results. In stove mode, it provides an at a glance measure of the temperature of all your pans so you can judge if the food might burn or the heat needs to be turned up, and it automatically times how long each different pan has been cooking for. If you leave the app running in the background it will even work in your pocket, so if you want to check how long your pasta has been boiling, you can simply open your app to tell you.

2023 Update – Back in Production!

We finalised the design work, and partnered with a contract electronics manufacturing company here in the UK to start production just in time for the covid pandemic to hit our supply chains. Luckily pre-pandemic we had sufficient components on order to do our first production run, which sold out quickly. Unfortunately lockdown induced semiconductor shortages then hit us hard just as we began scaling up and we had to pause manufacturing due to lack of parts. However supply chains are finally freeing up, and I am proud to say we are back in production!

Thank you for taking the time to learn about the development of Pippa. To be one of the first to own a Pippa please order one now.

From the Inventors of LeakBot

 

 

Recent Posts
Categories